Anthropic announced that it had shut down state-sponsored espionage operations that used its artificial intelligence models for surveillance. It said the activity detected between January and July 2026 originated in China, Iran and West Africa and targeted minorities, political dissidents and diaspora communities.
Dissidents and minorities targeted in Asia and Iran
In another case, a contractor working for Mali’s national security agencies used the Claude model to design core software that enabled an intelligence-gathering operation. Anthropic said these incidents showed artificial intelligence being used as a substitute for engineering workforces. It also said it had thwarted user attempts to design weapons, conduct biological research it described as questionable, and develop scams through dating apps.
Using Claude for espionage and model development
The company accused Chinese developers of using the Claude model deceptively to answer users’ queries while using the resulting data to improve their own models through a process known as “distillation”, in which one artificial intelligence model is used to train another.
Anthropic conducts artificial intelligence research and development in the United States and develops the Claude family of models.
It also adopts an approach called “constitutional AI”, which aims to regulate model behavior and align it with human values.
